I wouldn't say meaningful, as it implies understanding.

Perhaps a compromise on Appropriate might be more accurate?

Quote: Original post by ROBERTREAD1
I wouldn't say meaningful, as it implies understanding.

Not at all... the only understanding has to reside in the observer. This is why Game AI is different to regular AI... it only has to pass the 'illusion' test.

Timkin

Game AI usually involves Agent-based AI, if you can call it that. And the definition of an intelligent agent is one that responds the best way it can to the input it is given, or the situation its in. So, classic example, we can't say an agent is dumb for walking on the side walk if it just so happen to be hit from debris falling from 30,000 feet up from some airplane passing by. Of course, if it had knowledge of this event and didn't react to it, then it wouldn't be intelligent.

As for what is Advanced AI? Well, I don't think we even agree on what "regular" AI is. Yes, everyone says "well, to define AI, we must first agree on a definition of 'intelligence.'" However, Knuth once posed the question of what is "artificial?" But that's a totally different conversation.

Personally, I don't feel there's an actual discrepency between "advanced" AI and "regular" AI. The big difference is in how you use AI. You can take the best learning strategy in the world and implement it in a poor way and work on a problem it wasn't designed to solve and it would give you crappy results. On the other hand, a well implemented basic method can give you great results when used properly. AI is just another tool. Sort of like everyone has like a zillion tools in the tool shed. Every tool has its use and the same goes to "techniques" in AI. If you really wanted to classify them that badly, I'd personally say "advanced" AI is like your power tools while "regular" AI is like your handy screwdriver.

Quote: Original post by Timkin
[...]Not at all... the only understanding has to reside in the observer. This is why Game AI is different to regular AI... it only has to pass the 'illusion' test.

Timkin

I like chess AI systems far better than typical game AI systems, because they can (to an extent) 'explain' their actions. 'X would be a good move because it secures square Y, protects pieces Z, A, and B that would otherwise be threatened, and any other move leaves option Q R S open for the opponent'
In games where the AI acts as a player (for example, RTS games where the AI follows the rules, or deathmatch fps games like quake3 and unreal 2004) and not simply an opponent (FPS games like halo where the player and opponents are not on equal footing because the player is severely outnumbered), I would love a similar kind of display that made sense to me.

It could also make bot vs bot battles an interesting thing to watch. I can just imagine having a game like The Opera generating action movies (minus the unneeded parts like dialog =-) in realtime, with subtitles or overlays showing the tactics of each side with enough detail to induce the 'Wow! What an awesome idea!' kind of feeling that chess players can get from seeing a very well-thought out plan go into action and then again when it finally comes to fruition.

I think that, to get a "crown of AI" award next year, you'll probably need something like a multi-level planning system (i e, I want goal X, how do I get there?), an adaptive (Bayesian?) desires system, and a fair bit of good-old game-specific knowledge coded in (typically as primitives and weightings for the more general algorithms). Making this all run in real time on both console CPUs (CPUs the level of a Pentium with overclocking) and modern desktop CPUs (multi-core, out-of-order, cache-dependent) would be an interesting challenge.

Quote: Original post by Extrarius
'Wow! What an awesome idea!'

So are you suggesting that you'd consider an AI to be advanced if it could not only find the easily computable solutions to problems, but also the ones that aren't necessarily obvious to you?

Most AI techniques could be implemented to include an explanation phase. Indeed, this is often something you want during development (particularly for adaptive agents) so that designers can work out what the AI was thinking when it made a decision.

Quote: Original post by hplus0603
I think that, to get a "crown of AI" award next year, you'll probably need something like a multi-level planning system (i e, I want goal X, how do I get there?), an adaptive (Bayesian?) desires system, and a fair bit of good-old game-specific knowledge coded in (typically as primitives and weightings for the more general algorithms). Making this all run in real time on both console CPUs (CPUs the level of a Pentium with overclocking) and modern desktop CPUs (multi-core, out-of-order, cache-dependent) would be an interesting challenge.

That's probably doable. I'd say that within 5 years, we'd be wanting to see this sort of thing for multiple significant characters within a story, all acting to create a dynamic storyline while working toward designer-specified story goals.

Quote: Original post by Timkin
[...]So are you suggesting that you'd consider an AI to be advanced if it could not only find the easily computable solutions to problems, but also the ones that aren't necessarily obvious to you?

Most AI techniques could be implemented to include an explanation phase. Indeed, this is often something you want during development (particularly for adaptive agents) so that designers can work out what the AI was thinking when it made a decision.[...]

Not just nonobvious, but also indirect and somewhat complicated. Combining these three attributes means a plan will be 'difficult' to bring to fruition because many things could go wrong (which should result in a different plan that is less interesting but still strong), but when it does work it would be very fun and potentially surprising.

The best AI is the AI that does what is is ment to do without flaw; meaning basicly, if you want a (in real life) robot that cleans the house, if can clean the house perfectly then it is perfect. However, what frob said is the most on the money:

Quote: Origanal post by frob:
The "creme de la creme" AI right now is what will be called "regular" a year from now. Todays "regular" was the "creme de la creme" of the year before.

When testing AI in real life, there are three major checkpoints (for the current decade): 1) Self awareness; 2) must be able to talk to other people and not be recongized as a psuedo-human; 3) must be able to problem solve as well as a average human or better.

This is just in reallife though; for games it depends on the type of game and how difficult it is meant to be. Just because an AI is weak dosn't mean it is poorly built, it could be ment to as a easy enemy. As I begun with: The best AI is the AI that does what it was meant to do. I am kinda going in circles I guess.

As other people say, a good game AI to me would be one that can play an learn to play better using exactly the same information the human player is given.

In the case of FPS shooting, the AI should have to do pattern recognition to target the enemies, at least.

Quote: Original post by Servant of the Lord
When testing AI in real life, there are three major checkpoints (for the current decade): 1) Self awareness; 2) must be able to talk to other people and not be recongized as a psuedo-human; 3) must be able to problem solve as well as a average human or better.

...only if you're talking strong AI...